DeepUrbanMomentum: An Online Deep-Learning System for Short-Term Urban Mobility Prediction

• DOI: 10.1609/aaai.v32i1.11338
• Publication Date: 2022-06-24T21:33:07Z
• Authors: Renhe Jiang; Xuan Song; Zipei Fan; Tianqi Xia; Quanjun Chen; Satoshi Miyazawa; Ryosuke Shibasaki
• Keywords: Uncorrelated; Mobility model

Big human mobility data are being continuously generated through a variety of sources, some which can be treated and used as streaming for understanding predicting urban dynamics. With such data, the online prediction short-term at city level great significance transportation scheduling, regulation, emergency management. In particular, when big rare events or disasters happen, large earthquakes severe traffic accidents, people change their behaviors from routine activities. This means people's movements will almost uncorrelated with past movements. Therefore, in this study, we build an system called DeepUrbanMomentum to conduct next predictions by using (the limited steps of) currently observed data. A deep-learning architecture built recurrent neural networks is designed effectively model these highly complex sequential huge area. Experimental results demonstrate superior performance our proposed compared existing approaches. Lastly, apply real scenario that applicable world.